Practical guidelines for Production and Implementation in accordance with EBU R 128

Transcription

1 EBU TECH 3343 Practical guidelines for Production and Implementation in accordance with EBU R 128 Supplementary information for EBU R 128 Status: Version 2.0 Geneva August

2 * Page intentionally left blank. This document is paginated for two sided printing

3 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Conformance Notation This document contains both normative text and informative text. All text is normative except for that in the Introduction, any explicitly labelled as Informative or individual paragraphs that start with Note:. Normative text describes indispensable or mandatory elements. It contains the conformance keywords shall, should or may, defined as follows: Shall and shall not : Should and should not : Indicate requirements to be followed strictly and from which no deviation is permitted in order to conform to the document. Indicate that, among several possibilities, one is recommended as particularly suitable, without mentioning or excluding others. OR indicate that a certain course of action is preferred but not necessarily required. OR indicate that (in the negative form) a certain possibility or course of action is deprecated but not prohibited. May and need not Indicate a course of action permissible within the limits of the document. Default identifies mandatory (in phrases containing shall ) or recommended (in phrases containing should ) presets that can, optionally, be overwritten by user action or supplemented with other options in advanced applications. Mandatory defaults must be supported. The support of recommended defaults is preferred, but not necessarily required. Informative text is potentially helpful to the user, but it is not indispensable and it can be removed, changed or added editorially without affecting the normative text. Informative text does not contain any conformance keywords. A conformant implementation is one that includes all mandatory provisions ( shall ) and, if implemented, all recommended provisions ( should ) as described. A conformant implementation need not implement optional provisions ( may ) and need not implement them as described. 3

4 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 4

5 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Contents 1. Introduction EBU R 128, ITU-R BS Programme Loudness Loudness Range True Peak Level (TPL), Maximum Permitted TPL R 128 Logo General Concept of Loudness Normalisation Peak vs. Loudness Normalisation of the Signal vs. Metadata Using the Measure Loudness Range Climbing the True Peak Strategies for Loudness Normalisation Production, Post-Production Loudness Metering for Production and Post-Production Ready, Set (Levels), GO! Loudness Range for Production and Post-Production What to Measure in Production and Post-Production Signal-Independent vs. Anchor-Based Normalisation Low Frequency Effects (LFE) Channel File-Based Production and Playout Building Blocks Generic Loudness Levelling Strategies - Processing Metadata Programme Loudness Metadata Dynamic Range Control Metadata Downmix Coefficients Alignment of Signals in the Light of Loudness Normalisation Alignment Signal and Level Listening Level Implementation and Migration Generic Migration and Implementation Advice Points of Action for Migration and Implementation

6 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 10. Genre-specific Issues Commercials (Advertisements) and Trailers Music References Acknowledgements Although this document is the result of much collaborative work within the EBU s PLOUD group, it is the long-suffering chairman of this group, Florian Camerer, who has first written, collated, enriched and distilled the text into its publication form over many, many weeks and months of effort. Dedication This document is dedicated to two great audio engineers; Gerhard Stoll and Gerhard Steinke. 6

7 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Practical guidelines for Production and Implementation in accordance with EBU R 128 EBU Committee First Issued Revised Re-issued Technical Committee 2011 Aug Keywords: Audio, Loudness, normalisation, production, implementation. 1. Introduction This document describes in practical detail one of the most fundamental changes in the history of audio in broadcasting; the change of the levelling paradigm from peak normalisation to loudness normalisation. It cannot be emphasized enough that loudness metering and loudness normalisation signify a true audio levelling revolution. This change is vital because of the problem which has become a major source of irritation for television and radio audiences around the world; that of the jump in audio levels at the breaks in programmes, between programmes and between channels (see footnote 2 for a definition of programme ). The loudness-levelling paradigm affects all stages of an audio broadcast signal, from production to distribution and transmission. Thus, the ultimate goal is to harmonise audio loudness levels to achieve an equal universal loudness level for the benefit of the listener. It must be emphasised right away that this does not mean that the loudness level shall be all the time constant and uniform within a programme, on the contrary! Loudness normalisation shall ensure that the average loudness of the whole programme is the same for all programmes; within a programme the loudness level can of course vary according to artistic and technical needs. With a new (true) peak level and the (for most cases) lower average loudness level the possible dynamic range (or rather Loudness Range ; see 2.2) is actually greater than with current peak normalisation and mixing practices in broadcasting. The basis of the concept of loudness normalisation is a combination of EBU Technical Recommendation R 128 Loudness normalisation and permitted maximum level of audio signals [1] and Recommendation ITU-R BS.1770 Algorithms to measure audio programme loudness and true-peak audio level [2]. 7

8 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 In addition to R 128, the EBU PLOUD group has published four other documents: EBU Tech Doc 3341 Loudness Metering: EBU Mode metering to supplement loudness normalisation in accordance with EBU R 128 [3] EBU Tech Doc 3342 Loudness Range: A measure to supplement loudness normalisation in accordance with EBU R 128 [4] EBU Tech Doc 3343 Practical guidelines for Production and Implementation in accordance with EBU R 128 [this document] and EBU Tech Doc 3344 Practical guidelines for distribution systems in accordance with EBU R 128 [5] The Technical Documents about Loudness Metering and about the measure Loudness Range also play an important role for the practical implementation of loudness normalisation. They will be introduced as well and referred to in the relevant sections. The Distribution Guidelines close the circle, covering all aspects of loudness normalisation for the distribution of audio signals and addressing the critical links between production and the final recipient, the consumer. As this is a very detailed document in itself it will not be introduced here except for the occasional reference. At the beginning of these Practical Guidelines the core parts of EBU R 128 and ITU-R BS.1770 are introduced, followed by the general concept and philosophy of loudness normalisation. The document will then look at loudness strategies for production and post-production (metering, mixing, Metadata, etc.), and for file-based workflows, that is, ingest, playout and archiving issues (metering, automated normalisation, Metadata etc.). A separate chapter will look at Metadata in more detail. Alignment of audio signals and studio listening levels are discussed, and practical advice is given for the transition to loudnessnormalised production (implementation and migration). Genre-specific issues regarding commercials (advertisements) and trailers as well as music programmes will be addressed in the last chapter. These Practical Guidelines are meant to be a living document, where, over time, experiences of broadcasters will find its way into the document, providing additional information and guidance for this fundamental change of the way, audio signals are treated and balanced to each other. Please note that many standards documents are subject to revision from time to time, including this one. You are strongly advised to check for the latest versions. 8

9 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R EBU R 128, ITU-R BS.1770 EBU R 128 establishes a predictable and well-defined method to measure the loudness level for news, sports, advertisements, drama, music, promotions, film etc. throughout the broadcast chain and thereby helps professionals to create robust specifications for ingest, production, play-out and distribution to a multitude of platforms. R 128 is based entirely on open standards and aims to harmonise the way we produce and measure audio internationally. The basis of R 128 is ITU-R BS.1770, the result of extensive work by the International Telecommunication Union. The purpose of that standard was to establish an agreed open algorithm for the measurement of loudness and the true peak levels of programmes. It is a robust standard which has the benefit of a simple implementation. In brief, it defines a K-weighting filter curve (a modified second-order high-pass filter, see Figure 1) which forms the basis for matching an inherently subjective impression with an objective measurement. Figure 1: K-Weighting filter curve for loudness measurement This weighting curve is applied to all channels (except the Low-Frequency Effects Channel (LFE) which is currently discarded from the measurement; see below), the total mean square level is calculated (with different gain factors for the front and surround channels; see Figure 2) and the result is displayed as LKFS (Loudness, K-Weighting, referenced to digital Full Scale), or LUFS 1 (Loudness Units, referenced to digital Full Scale). For relative measurements, Loudness Units (LU) are used, where 1 LU is equivalent to 1 db. 1 The EBU recommends the use of LUFS (as specified in EBU Tech Doc 3341). LUFS is equivalent to LKFS and overcomes an inconsistency between ITU-R BS.1770 and ITU-R BS LUFS also complies with the international naming standard ISO [6]. 9

10 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Low Frequency Effects (LFE) channel The Low Frequency Effects channel (the.1 -channel in 5.1 ) of a multichannel audio signal is currently not taken into account for the loudness measurement according to ITU-R BS This may lead to abuse of the LFE with unnecessary high signal levels. Ongoing investigations try to evaluate the subjective effect the LFE has on the perception of loudness as well as the appropriate way to include it in the objective loudness measurement. Figure 2: Channel processing and summation in ITU-R BS.1770 Whereas BS.1770 defines the measurement method, R 128 extended it by actually defining a specific Target Level for loudness normalisation as well as a gating method to improve the loudness matching of programmes which contain longer periods of silence or isolated utterances. The EBU's development was needed to accommodate the needs of programme makers, with particular regard to having a means to measure complete mixes (rather than just one component, such as speech or music) and the loudness range of the programme. 10

11 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 To do this, the EBU has specified three new measures: Programme Loudness Loudness Range True Peak Level 2.1 Programme Loudness Programme Loudness describes the long-term integrated loudness over the duration of a programme 2. The parameter consists of one number (in LUFS, with one number after the decimal point) which indicates how loud the programme is on average. This is measured with a meter compliant with ITU-R BS.1770 including a gating function. The gate serves to pause the loudness measurement when the signal drops below a certain threshold. Without this gating function, programmes with longer periods of silence, low-level background sounds or noise will get too low an integrated loudness level value. Such programmes would subsequently be too loud after normalisation. Following a series of tests, a gate of -8 LU (Loudness Units; 1 LU 1 db) relative to the ungated LUFS measurement was agreed to by the EBU (see [7] for more details). Subsequently, the suggestion to include the relative gating method into BS.1770 was sent to the ITU. In ITU-R BS , the relative-threshold gate is now part of the measurement algorithm for integrated loudness, albeit with the slightly lower threshold of -10 LU. EBU R 128 and all supporting documents adopt this new threshold. (Details of the gating function need only concern manufacturers. From the user's perspective there will be little difference, but users are advised to keep all their equipment up-to-date to ensure consistency of measurement.) The listening tests also confirmed, along with other findings, the choice of the Target Level to which every audio signal will be normalised; it is: LUFS (-10 rel gate) Why -23 LUFS? Investigations and measurements of the loudness level of actual broadcasting stations showed an average loudness level of about -20 LUFS (with lots of outliers...). Another input to that discussion came from the ITU, in the form of the document ITU-R BS.1864 Operational practices for loudness 2 The term a programme is also used to mean an advertisement, a promotional item etc. For clarity, the advertisements etc. which are placed around and within the running time of what is generally considered to be 'a programme' are treated as programmes in their own right (also individual advertisements within a block are separate programmes); their integration with the longer programmes is thus made easier. Evidently, the makers of either type of programme can have no knowledge of what will be placed with it and so each type has to be considered separately. In this document, the term programme refers to the programme as completed by Production and not the combination of the programme, interstitials, and advertisements that arrives at the viewer's or listener's receiver within the overall running time of the programme. 11

12 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 in the international exchange of digital television programmes [8]. In BS.1864, a Target Level of -24 LUFS is recommended, although without anticipating the inclusion of a gating function in BS Informal tests conducted by members of the EBU PLOUD group have shown that the difference in the loudness measurement with and without the -10 LU relative gate of programmes with a small to medium loudness range are around 0 1 LU. -23 LUFS with the gate is therefore in many cases almost equivalent to -24 LUFS without a gating method. -23 LUFS (with the gate) was therefore considered to be the lowest possible programme loudness level without making the transition from an average level of -20 LUFS even more challenging. As -20 LUFS was thought to allow not enough headroom for dynamic content, the decision was taken to settle on LUFS. A deviation of ±1.0 LU is acceptable for programmes where an exact normalisation to the Target Level of LUFS is not achievable practically (such as live programmes or ones which have an exceedingly short turn-round). In cases where the levels of a programme s individual signals are to a large extent unpredictable, where a programme consists of only background elements (for example, the music bed for a weather programme), or where programmes are deliberately mixed lower, the programme loudness level may lie outside the tolerance. Such cases should be increasingly rare, though. 2.2 Loudness Range Another major topic was the loudness range which would be needed to accommodate all programmes (provided that they don t exceed the tolerable loudness range for domestic listening). The measure Loudness Range (LRA) quantifies (in LU) the variation of the loudness measurement of a programme. It is based on the statistical distribution of loudness within a programme, thereby excluding the extremes. Therefore, for example, a single gunshot is not able to bias the outcome of the LRA computation. EBU Recommendation R 128 does not specify a maximum permitted LRA, as it is dependent on factors such as the tolerance window of the average listener to the station, the distribution of genres of the station etc. R 128 does, however, strongly encourage the use of LRA to determine if dynamic treatment of an audio signal is needed and to match the signal with the requirements of a particular transmission channel or platform. More details about LRA may be found in EBU Tech Doc First experiences at broadcasting stations suggest a maximum LRA value of approximately 20 LU for highly dynamic material, such as action movies or classical music. The majority of programming will never need to fully use such a high LRA value or, indeed, be able to reach it! For very short programmes (<30 s) such as commercials, advertisements or trailers, setting a limit for the maximum values of the Momentary or Short-term Loudness Level 3 may provide a better way to control the dynamic properties as a kind of second line of defence (see 7, 10). 3 Maximum Momentary Loudness Level (Max ML) is the highest value (in LUFS) of an audio signal s Momentary Loudness Level (integration time 400 ms). Maximum Short-term Loudness Level (Max SL) is the highest value (in LUFS) of an audio signal s Short-term Loudness Level (integration time 3 s). 12

13 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Figure 3: Loudness Range (LRA) as a result of the statistical distribution of loudness levels Figure 3 shows the loudness distribution and LRA of the movie The Matrix ; 25 LU is probably challenging for most living rooms True Peak Level (TPL), Maximum Permitted TPL In Europe, the most widespread metering device was (and still is to a large extent) the Quasi Peak Programme Meter (QPPM; integration time = 10 ms). With the transition to digital signal processing, sample peak meters appeared. While a QPPM cannot display short peaks (<<10 ms) by design, also a sample peak measurement may not reveal the actual peak level represented by a digital signal. Digital processing or lossy coding can cause inter-sample peaks that exceed the indicated sample level. In broadcasting it is important to have a reliable indication of level across platforms and across sample rates. This meter should indicate clipping, even when the peak lies in between samples, so that the distortion that can happen in subsequent Digital-to-Analogue converters, sample rate converters or commonly used codecs can be predicted and avoided. A sample peak meter cannot do that and is therefore insufficient for use in modern broadcasting (see Lund, Th.: 'Stop counting samples' [9]). The true peak level indicates the maximum (positive or negative) value of the signal waveform in the continuous time domain; this value may be higher than the largest sample value in the time-sampled domain. With an oversampling true-peak meter compliant with ITU-R BS.1770, those true peaks (unit symbol according to ITU-R BS.1770: dbtp decibel referenced to digital Full Scale measured with a True Peak meter) are now able to be detected. The accuracy depends on the 13

14 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 oversampling frequency. It is only necessary to leave a headroom of 1 db below 0 dbfs to still accommodate the potential under-read of about 0.5 db (for a 4x oversampling true-peak meter; basic sample rate: 48 khz). The Maximum Permitted True Peak Level recommended in R 128 will consequently be: -1 dbtp This is applicable to the production environment for generic linear audio signals. Note that some parts of the chain, such as analogue re-broadcasters and users of commonly used data reduction codecs require a lower True Peak Level. The EBU Distribution Guidelines (EBU Tech Doc 3344) contain comprehensive coverage of the topic. Summary of EBU R 128 The measures Programme Loudness, Loudness Range and Maximum True Peak Level characterise an audio signal; The Programme Loudness Level shall be normalised to LUFS; The tolerance is generally ±1.0 LU for programmes where an exact normalisation is not achievable practically; The measurement shall be done with a meter compliant with ITU-R BS and EBU Tech Doc 3341 ( EBU mode amongst other things summarising the gating method described in BS ); The measure Loudness Range shall be used to help decide if dynamic compression is needed (dependent on genre, target audience and transmission platform); The Maximum Permitted True Peak Level in production is -1 dbtp; Loudness Metadata shall be set to indicate -23 LUFS (for programmes that have been normalised to that level, as is recommended); loudness Metadata shall always indicate the correct value for programme loudness even if for any reason a programme may not be normalised to -23 LUFS. 14

15 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R R 128 Logo The EBU has introduced an official logo for R 128, comprised of the numbers 1, 2 and 8 forming a happy, smiling face: The logo may be used (with certain prerequisites) by manufacturers to indicate compliance with EBU Mode. 3. General Concept of Loudness Normalisation 3.1 Peak vs. Loudness The still widespread audio levelling concept of peak normalisation with reference to a Permitted Maximum Level (PML; for example, -9 dbfs), has led to uniform peak levels of programmes, but widely varying loudness levels. The actual variation is dependent on the degree of dynamic compression of the signal. In contrast, loudness normalisation achieves equal average loudness of programmes with the peaks varying depending on the content as well as on the artistic and technical needs (see Figure 4). Provided the loudness range of a programme is within the permitted tolerance, the listener can enjoy a uniform average loudness level over all programmes, thus not having to use the remote control for frequent volume adjustments any more. 15

16 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Figure 4: Peak level normalisation vs. Loudness level normalisation of a series of programmes Again, this does NOT mean that within a programme the loudness level has to be constant, on the contrary! It also does NOT mean that individual components of a programme (for example, pre-mixes or stem-mixes, a Music & Effects version or an isolated voice-over track) have all to be at the same loudness level! Loudness variation is an artistic tool, and the concept of loudness normalisation according to R 128 actually encourages more dynamic mixing! It is the average, integrated loudness of the whole programme that is normalised. 3.2 Normalisation of the Signal vs. Metadata There are basically two ways to achieve loudness normalisation for the consumer: one is the actual normalisation of the audio signal itself, so that the programmes are equally loud by design the other method is with the use of Loudness Metadata that describes how loud a programme is. For the latter, the actual average programme loudness levels don t need to be changed to a normalised value and can still vary a lot from programme to programme. For those with up-to-date equipment, the normalisation can be performed at the consumer s end using the individual loudness Metadata values to gain-range the programmes to the same replay level. 16

17 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Within the EBU R 128 loudness levelling paradigm the first solution is encouraged due to the following advantages: simplicity and potential quality gain of the audio signal. The second solution is not forbidden (see also the Distribution Guidelines document, EBU Tech Doc 3344), but having one single number (-23 LUFS) has great strength in spreading the loudness-levelling concept, as it is easy to understand and act upon. And the active normalisation of the source in a way punishes overcompressed signals and thus automatically encourages production people to think about other, more dynamic and creative ways to make an impact with their programme. In other words, the actual technical change of the audio signal level through active normalisation to -23 LUFS has direct consequences on the artistic process and in a positive way! The production side is thus relieved from fighting the loudness war an unfortunate and widespread result of the peak-normalisation paradigm. Nevertheless, it must be stated that both methods can complement each other, they are not to be seen as opponents or a black-and-white view of the same issue. Both approaches are a part of R 128 but because of the advantages listed above, the normalisation of the audio signal is recommended. Working towards a common loudness level signifies a whole new concept of mixing, of levelling, of generally working with audio. Whereas a peak limiter set to the Permitted Maximum Level (usually -9 dbfs, measured with a QPPM) provided a sort of safety ceiling where, no matter how hard you hit it, it always ensured the correct maximum level, the loudness levelling paradigm more resembles floating in space, with the open sky above (see Figure 5). 17

18 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Figure 5: Quasi-Peak Level normalisation ( safety ceiling ) vs. Loudness Level normalisation With loudness normalisation and metering, the safety ceiling is gone. This might be intimidating for some, as it was in a way comfortable that one didn t have to watch levels so closely the limiter at the end of the chain ensured that your output was always tamed. But the side effect was that loudness levels went up, because more sophisticated processors exhibited fewer dynamic compression and limiting artefacts. Loudness levelling, on the other hand, encourages the use of by far the best metering device: the ear. This implies more alert mixing and fosters audio quality. Experience of several EBU members has shown that working with the loudness paradigm is liberating and satisfactory. The fight for Who is the loudest is gone, overall levels go down, and this in combination with a higher Maximum Permitted True Peak Level (-1 dbtp) results in potentially more dynamic mixes with greater consistency of loudness within a programme. Dynamic compression is again an artistic tool and not a loudness weapon, the audio quality increases! Putting mixing by ear back on track is a welcome relief and long overdue. The mixer is now encouraged to mix by ear alone (another effect of loudness metering) after setting levels at a fixed monitor volume (see 8). Downstream of production, the broadcaster is confronted with the need to normalise diverse content originating from different places. Especially during the transition period there will still be 18

19 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 many programmes that are not yet loudness normalised. Strategies for these programmes have to be developed, like automated normalisation directly after ingest to a playout server or the installation of a safety loudness regulation device at the output of master control to be able to handle, for example, live feeds that are not produced to the target level of -23 LUFS. Issues like these will be covered with additional details later on ( 4.3, 6). 3.3 Using the Measure Loudness Range With Loudness Range (LRA) it is now possible to quantify the dynamics of a programme. In the past, it had to be educated guesswork of experienced audio personnel to decide if a programme would fit into the loudness tolerance window of the intended audience. Using Loudness Range, the guesswork is over at the end of the measurement period (usually the whole programme), a single number enables the mixer/operator to decide if further dynamic treatment is necessary. It is important to understand that it is impossible to define one maximum value of LRA for all broadcasters and all programmes. The individual maximum LRA is dependent on the genre(s) (theme channels with very uniform content like news will certainly not have as high a maximum LRA as public broadcasters with a great variety of genres for programmes like, for example, a classical music concert). The maximum LRA value may also be different for distribution platforms like mobile broadcasting as well as different replay environments (see Figure 6; the distance between the yellow lines indicates examples of different Loudness Range values). The average listening environment, age of the target audience, listening comfort zone of the consumer and other parameters all influence the choice of a station s maximum LRA values for specific programming. The Loudness Range Control Paradigm starts from a generic maximum accepted value of Loudness Range according to the principles described above and adapts this value downstream to comply with technical necessities of individual distribution platforms and replay environments. Figure 6: Different examples for Loudness Range depending on the replay environment It is the responsibility of the mixer at the production stage of the programme to determine the programme's LRA; the mixer is advised to respect the paradigm. In environments where human control and intervention are not possible or in place, the measurement of LRA may lead to the use of suitable presets of a dynamic processor that has been configured for the individual genre of the 19

20 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 audio signal. Nevertheless, it is advisable to strive for a situation where an audio engineer can influence LRA according to the specifications of the broadcaster, as this potentially increases audio quality. As a result of the need for different values of Loudness Range, EBU R 128 does not include a maximum permitted LRA value, but instead strongly encourages the use of the Loudness Range measure to evaluate the potential need for dynamic range processing according to the different criteria mentioned above. Loudness Range is also a useful indicator of potential dynamics reduction processes in a signal chain, performed on purpose or accidentally. If the LRA value of a programme after it has passed through a processor chain is, for example, lower than it was originally, such a reduction process has occurred. 3.4 Climbing the True Peak The third measure recommended by R 128 concerns the maximum true peak level of an audio signal. Having abandoned the peak normalisation paradigm, it is of course still vital to measure and control the peaks of a programme, and especially its maximum peak to avoid overload and distortion. A loudness meter compliant with EBU Mode (see EBU Tech Doc 3341) also features the measurement and display of the true peak levels of a programme. Safety limiters to avoid overmodulation will have to be able to work in true-peak mode and need to be adjusted to the appropriate Maximum Permitted True Peak Level, in production as well as at the output of master control, at the distribution headend and the transmitter site. Next to the Maximum True Peak Level for generic PCM signals in production (-1 dbtp), further suggestions for different applications are given in EBU Tech Doc 3344 ( Distribution Guidelines ). 20

21 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Strategies for Loudness Normalisation 4.1 Production, Post-Production Approaching loudness normalisation in these areas offers two possibilities: the first is to keep current levelling practices and perform a level shift afterwards, and the second is to change the levelling habit to loudness control and normalisation with no or only a small shift needed (Figure 7). The first approach is more relevant for the early stages of the transition, and it is perhaps especially useful to those who work on live programmes. The existing meters, limiters and mixing practices are retained and a level shift is done at the output of the console (after the main meters) to achieve the loudness Target Level of -23 LUFS. Figure 7: Two principal working methods to achieve uniform loudness in production and post-production A loudness meter is placed after the level shift to enable the engineers to understand the exact amount of shift (which initially is still a bit of guesswork). Using a loudness meter in parallel with a conventional meter is in any case a good idea. In this way experience can be gained before actually diving into the loudness-levelling world. Furthermore, using a loudness meter to measure past programmes of the same genre gives good guidance as to where the levels sit. For programmes that are finished in post-production the necessary level shift for approach 1 is easy to perform. Measuring the whole programme in one go, the necessary gain offset can be determined exactly, and in today s file-based world a gain calculation is a very quick and easy operation. Of course, for live programmes it is challenging (if not a matter of luck) to exactly achieve Target Level. Therefore, a deviation of ±1.0 LU is acceptable for programmes where an exact normalisation to the Target Level of -23 LUFS is not achievable practically (in addition to live programmes, for example, ones which have an exceedingly short turn-round). Early experience at NDR (North-German Broadcasting), ORF (Austrian Broadcasting), and RTBF (Belgian Broadcasting French part) has shown that it is certainly possible for live mixes to fall within the ±1 LU window permitted by EBU R

22 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 In cases where the levels of a programme s individual signals are to a large extent unpredictable, where a programme deliberately consists of only background elements (for example, the music bed for a weather programme) or where the dramaturgical intention of a programme makes a loudness level particularly lower than target level desirable, this tolerance may be too tight. It is therefore anticipated for such cases that the integrated loudness level may lie outside the tolerance specified in R 128. For levelling solution 1 (keeping current levelling practices) it is likely that in almost all cases the necessary gain shift will be negative (attenuation). Therefore an additional step of reducing the dynamic range and/or limiting the Maximum True Peak Level is usually not necessary. The potential attenuation in the vast majority of cases is also the reason why the Metadata solution described in 3.2 is not advised for solution 1. Levelling solution 2 (changing to loudness normalisation right away) is the one that is recommended in these Practical Guidelines. Again, after an initial measurement and testing period of past programmes and the installation of a loudness meter in parallel to the meter currently used (usually a QPPM), the advantages of the loudness-levelling paradigm speak for themselves. The greater dynamic range possible will be a welcome bonus for crowd noise, for example, of sports programmes, enhancing the impact of a game for the viewers and listeners. Studio voice-overs that are often dynamically compressed due to artistic reasons (and where therefore the loudness-to-peak ratio will be lower) will be better balanced with more dynamic original location recordings etc. etc. In what follows, the impact of working with a loudness meter in production and post-production will be examined. 22

23 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Loudness Metering for Production and Post-Production An EBU Mode loudness meter as defined in EBU Tech Doc 3341 offers 3 distinct time scales: Momentary Loudness (abbreviated M ) time window: 400 ms Short-term Loudness (abbreviated S ) time window: 3 s Integrated Loudness (abbreviated I ) from start to stop The M and S time windows 4 are intended to be used for the immediate levelling and mixing of audio signals. Initial level setting may be performed best with the Momentary Loudness Meter, adjusting the level of key or anchor elements (such as voice, music or sound effects) to be around the Target Level of -23 LUFS. Of course a mixer has to know at any time how loud the actual signal is, and that is the main purpose of the Momentary and Short-term measurement (both are not gated). Due to an inconsistency between ITU-R BS.1770 and ITU-R different naming convention, complying with ISO : BS.1771, EBU Tech Doc 3341 suggests a The symbol for Loudness Level, K-weighted should be L K. The unit symbol LUFS indicates the value of L K with reference to digital full scale. The unit symbol LU indicates L K without a direct absolute reference and thus also describes loudness level differences. Any graphical or user-interface details of a loudness meter complying with EBU Mode have deliberately not been specified; nevertheless, two scales have been defined: EBU +9 Scale which ought to be suitable for most programmes and EBU +18 Scale which may be needed for programmes with a wide LRA. Both scales can either display the relative Loudness Level in LU, or the absolute one in LUFS. 0 LU in EBU mode equals the target level of -23 LUFS. The meter manufacturers in the PLOUD Group have agreed to implement the EBU Mode set of parameters, to make sure their meters readings will be aligned. Many more manufacturers have adopted EBU Mode too, or are in the process of doing so. Figure 8 shows a schematic representation of a bar-graph meter with the two EBU-mode scales; Figure 9 shows how a software meter based on a needle could look. 4 M and S are commonly used in stereophony for Mid and Side. To distinguish the integration times Momentary and Short-term, the versions ML K and SL K (as well as IL K ) may be used. L K stands for Loudness Level, K- weighted, and complies with the international naming standard ISO

24 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Figure 8: A schematic representation of the two loudness scales (here in LU) as described in EBU Tech Doc 3341 Figure 9: A schematic representation of an emulated loudness meter with a bendy needle 4.3 Ready, Set (Levels), GO! It is advisable to set levels with a bit of caution initially, as it is psychologically easier to gradually increase your integrated loudness level during a mix than to decrease it. The gating function also has an influence: bigger level changes are needed to bring the average level down than to raise it. Usually, a slight increase in the course of a programme is also dramaturgically more natural and an initially defensive strategy leaves the engineer room to manoeuvre in case of unexpected or unpredictable signals and events. 24

25 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Once levels of individual signals are set, and a fixed monitor gain has been established (see 8), the audio engineer can switch to mixing only by ear. Watching the Momentary or Short-term loudness level and an occasional glance at the value of the Integrated loudness level should give enough confirmation that the mix is on the right track towards Target Level. With a numerical readout of the I -value with one decimal point precision or a graphical display of similar resolution, trends can be anticipated and the appropriate measures taken. This should be performed in a smooth manner, as too drastic changes will, in most cases, be artistically unsatisfactory. With the Maximum Permitted True Peak Level being -1 dbtp, the phenomenon of hitting the wall (meaning the safety limiter operating at -9 dbfs) is now less likely to occur. Used reasonably and with a clear intention, this opening of the lid together with loudness normalisation to -23 LUFS results in more dynamic mixes, in less dynamic compression artefacts like pumping and thus in an overall increase of audio quality! Programme makers who favoured dynamic mixes in the past are now relieved from potential compromises because their programme would sound softer than more compressed ones. With loudness normalisation, this compromise is gone. At last! The elements of a mix that are most important for a uniform subjective loudness impression are so-called foreground sounds like voice, music or key sound effects. Individual sound elements do have a widely varying difference between their loudness level and their peak level. For example, the clink of two glasses when toasting has a high peak level, but quite low loudness level. On the other hand, a dynamically compressed hard rock guitar riff has a loudness level that is almost the same as its peak level! If those two signals are aligned according to their peaks, the guitar riff will be much louder than the clink of the glasses. This example is meant to illustrate the concept, it does NOT mean that those two signals are necessarily to be mixed with equal loudness! The level of individual elements and components (like pre-mixes or stem-mixes, a music-only mix or a voice-over track) in the mix is an artistic decision, naturally, but loudness metering can help the mixer with useful visual feedback that actually shows what he or she hears! Coming back to metering, at the end of a programme there are two scenarios: having exactly hit target level (-23.0 LUFS) or having missed target level in either direction For live productions, understandably the second scenario will be more likely. If the actual loudness level is within the accepted tolerance of ±1.0 LU, then no further action is needed. If the level lies outside this tolerance due to the particularly unpredictable nature of the programme or the rare occurrence of foreground elements, this is still acceptable from a generic production standpoint (as mentioned earlier). Measures may be taken further downstream to tame these cases in the form of loudness processors that gradually adjust the integrated loudness level of such programmes in an unobtrusive manner and can act as a sort of loudness safety net. This must be achieved with a reasonably slow reaction time, so that the inner dynamics of the production are not harmed. Differentiation between live and file-based programmes should be possible as far as the individual preset of such a loudness processor is concerned or where in the signal chain the processor is installed. The processor may only be needed for live programmes if the workflow for file-based programmes is already fully compliant with EBU R 128. If a downstream dynamics and loudness processor is situated at the output of the Master Control Room, it should be able to be bypassed for programmes compliant with R 128. This bypass situation is expected to become the normal way of working the more programmes are on target, as the ultimate recommended goal is to tackle and normalise the audio signal itself. 25

26 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 In the post-production area one is more likely to hit target level because of the very nature of the workflow with more opportunities to redo and change the mix and thus the loudness levels. Furthermore, usually there is enough time to perform a complete integrated measurement of the whole programme once it is finished, as well as a subsequent gain correction. In a file-based production environment this correction can be performed much faster than realtime. Situations may be common where mixes in the post-production area are performed as live, that is, for example, directly onto a tape with few if any mistakes from voice talents (in case of a voice-over mix). Also, for example, a 1:1 tape-copy process with loudness adjustment on-the-fly falls into this category. These situations are then similar to live productions and should be approached accordingly. Especially in the transition phase moving towards loudness normalisation such aforementioned loudness processors downstream will certainly be helpful for broadcasters to adapt to the loudness levelling system and catching possible outliers. It should be the goal of the broadcaster (and also the mixing engineers) to have these processors work as little as possible, as the integrated loudness level of programmes is increasingly within the accepted loudness tolerance. The exact transition scenario, time scale and implementation plan are of course individually different for each broadcaster (see 9). While this is anticipated, in the interest of the consumer the switch to loudness normalisation should be performed in due time as the benefit for the listener is so substantial. 4.4 Loudness Range for Production and Post-Production Working with loudness normalisation right away also implies controlling Loudness Range (LRA) as the dynamic possibilities are expanded. This is important to ensure an appropriate signal for the intended audience and distribution chain. Whereas in production and post-production a generic mix may be created (with a relatively high LRA value and a Maximum Permitted True Peak Level of -1 dbtp), different platforms may need a lower LRA value and a lower Maximum Permitted True Peak Level (while keeping the Programme Loudness Level at -23 LUFS). The system within R 128 appreciates this generic approach with further processing downstream to tailor the signal to individual environments and platforms. With the measure Loudness Range it is now systematically possible to determine appropriate measures for potential dynamic compression of a programme to fit it to the tolerance window of the audience or distribution platform. In practice, overall low-level compression may lead to satisfactory results (see Figure 10 as an example): a low threshold (< -40 dbfs) and a moderate compression ratio (1:1.2 1:1.5), together with a long release-time (>1 s), ensure uniform compression of the whole signal range. Dependent on the original loudness level, a shift to the Target Level of -23 LUFS may be performed in parallel through adjusting the make-up gain of the compressor accordingly. 26

27 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 Figure 10: Example for processing of Loudness Range (LRA) with a compressor with a low threshold (-50 dbfs) and a moderate compression ratio (1:1.5) 27

28 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 5. What to Measure in Production and Post-Production 5.1 Signal-Independent vs. Anchor-Based Normalisation EBU R 128 recommends measuring the whole programme, independent of individual signal types like voice, music or sound effects (see figure 11). This is considered to be the most generally applicable practice for the vast majority of programmes: Figure 11: Elements of a programme For programmes with an increasingly wide loudness range (>12 LU, approximately) one may optionally use a so-called anchor signal for loudness normalisation, thus performing an individual gating method, so to speak. This should be a signal which the producer or engineer wants to be representative for the average loudness of the programme, like speech or a singing voice, a certain part of a music programme in mezzoforte, a consistently applied and dramaturgically important sound effect sequence etc. It must be emphasized, though, that choosing an anchor signal is an active process requiring input from an experienced operator. This approach should only be considered after operators and sound engineers have become very comfortable with the concept of loudness normalisation. Performed well, it may help to fine-tune the loudness of wide loudness range programmes according to the chosen anchor signal. There also exists an automatic measurement of one specific anchor signal in the form of Dialogue Intelligence, a proprietary algorithm of Dolby Laboratories, anticipating that speech is a common and important signal in broadcasting. The algorithm detects if speech is present in a programme and, when activated, only measures the loudness during the speech intervals. For programmes with a narrow loudness range the difference between a measurement restricted to speech and one performed on the whole programme is small, usually <1 LU. For programmes with a wide loudness range, such as action movies, this difference gets potentially bigger, sometimes exceeding 4 LU. Automatic detection of an anchor signal is intended to help identify what should be at Target Level. Like any algorithm for detecting specific signals out of a complete and complex mix, speech discrimination can be tricked either by signals closely resembling the spectral pattern of speech (for example, certain woodwind instruments or a solo violin) or by speech signals that are too far off the discrimination threshold (for example, certain language dialects). For programmes where these anchor signals are consistently moving around the discrimination threshold, the loudness measurement can also vary significantly if the measurement is performed repeatedly. 28

29 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 For short programmes like commercials, advertisements, trailers and promotional items, (automatic) speech normalisation is likely to give unsatisfactory results in the light of the future increase of loudness range and potentially enhanced dramaturgical concepts. In such cases, most international recommendations (also this one) agree on measuring all by all means. In any case, broadcasters have to be aware that especially in a file-based environment, where for most content the whole programme independent of signal type (speech, music, sound effects) will be measured automatically, a different strategy might have to be established to treat programmes based on anchor normalisation. To summarise: It is because of these uncertainties and the fact that speech represents only one part of the whole programme (albeit a very important and common one) that R 128 recommends measuring all that is the whole programme, independent on the signal type (such as voice, music or sound effects). This is supported by the following observations: The difference between measuring all and measuring an anchor signal (such as voice, music or sound effects) is small for programmes with a narrow Loudness Range; The difference between all and anchor measurements depends strongly on the content of the programme, but can be expected to be bigger if the Loudness Range is bigger; Automatic anchor signal discrimination may perform well for a majority of programmes, but may be tricked by similar signals or may not trigger at all, thus not giving 100% consistent results; File-based environments need a measurement paradigm that is applicable to 100% of the content and that delivers results that are good enough for all programmes; Identifying an anchor signal needs input from an experienced operator or a discrimination algorithm; such an algorithm may be subject to the potential uncertainties listed above. Anchor normalisation could offer better results on wide LRA material. It is however a task requiring expertise, and thus time and money, and if automatic discrimination is used, such an algorithm cannot be 100% reliable. Special measures need to be taken when anchor-adjusted content enters normalisation systems on file servers and thus needs bypassing the automatic processes in place. As the biggest common denominator, R 128 recommends to measure the whole programme with all its elements instead of anchors, even with wide LRA material. 5.2 Low Frequency Effects (LFE) Channel As noted in the description of ITU-R BS.1770 (see 2), the LFE channel is currently excluded from the measurement. One of the reasons is the widespread uncertainty of consumers and audio engineers as well as equipment implementation differences regarding the alignment of this channel (+10 db in-band gain). The omission of the LFE channel during the loudness measurement might cause its abuse. Further investigations of this matter and practical experience are needed to decide if and in which way the LFE signal might be included. One solution to completely avoid all potential issues with the LFE signal is not to use it at all, for example, if there is no need for extra headroom in the low bass region. 29

30 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 6. File-Based Production and Playout As the broadcast world is changing to file-based workflows, it is vital that the loudness normalisation concept is also fully embraced there. The basic principle stays the same: loudness normalisation and dynamic control of the audio signal are recommended, especially for new content. Nevertheless, as Metadata is an integral part of file-based systems, solutions that rely more on Metadata are described as well ( 7). The origin of a broadcast file that contains audio signals can be via an ingest process, via transfer from an external server and from a file-based archive. For existing programmes (archival content) there are basically four options to achieve loudness normalisation: Actually changing the the loudness level of all audio files to be on target Changing the loudness level only on demand Using the result of a loudness level measurement to adjust the playout level without changing the original loudness level Transporting the correct loudness Metadata to the consumer where normalisation is performed Which solution is actually chosen depends on factors like specific infrastructure, workflows, media asset management, availability of suitable equipment, financial resources, time etc. At the very beginning of a file s life inside a facility, measurements have to be made, providing the values for Programme Loudness Level, Loudness Range and Maximum True Peak Level the three characteristic audio measures defined in EBU R 128 (Maximum Momentary and Maximum Short-term Loudness Level may be measured and stored too for very short content (<30 s, see 7)). Depending on the results of that measurement and the subsequent method to achieve loudness normalisation and compliance with the acceptable Loudness Range, a processing scheme is executed, consisting of building blocks or core tasks. The workflow will now be examined in detail, with the help of generic flow-charts. 30

31 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Building Blocks Programme Loudness Level Processing (Figure 12) Figure 12: Programme Loudness processing block Loudness Range Processing (Figure 13) Figure 13: Loudness Range processing block 31

32 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Maximum True Peak Level Processing (Figure 14) Figure 14: Maximum True Peak Level processing block 6.2 Generic Loudness Levelling Strategies - Processing The three basic building blocks described above are at the core of any file quality control process regarding the technical parameters of its audio content. At the beginning of any potential processing the values of Programme Loudness Level (L K ), Loudness Range (LRA) and Maximum True Peak Level (Max TP) are measured. The result of this initial measurement determines the subsequent processing. Several different scenarios are possible: a) All three parameters are OK. This is obviously the ideal outcome of the measurement: the Programme Loudness Level is LUFS, the Loudness Range is within the specified limits of the broadcaster (depending on the genre and/or the distribution platform) and the Maximum True Peak Level is equal to or below the specified maximum value for the designated distribution system. 32

33 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 b) The Programme Loudness Level is higher than LUFS. A simple gain ranging (level reduction) operation solves that: Gain (db) = L K Target - L K measured (Example: the measured L K is LUFS; Target Level is LUFS; the necessary gain is [ (-19.4) =] -3.6 db. Max TP is naturally reduced by the same amount as L K.) c) The Programme Loudness Level is lower than LUFS. After applying a positive gain offset, the Maximum True Peak Level has to be recalculated (originally measured Max TP + gain offset = resulting Max TP) as it potentially lies above the permitted limit. If the new Max TP indeed exceeds the permitted limit, True Peak Limiting has to be performed according to the True Peak Processing Building Block. Another solution, which is applicable if such True Peak Limiting is not possible or wanted (or potentially too severe) is to leave L K at the original lower level and apply the appropriate loudness Metadata setting (lower than -23, reflecting the original loudness level). This requires that a fully functional system that supports and transports Metadata is in place (e.g. Dolby Digital, or MPEG-4). For both scenarios b and c a simple gain value stored as Metadata may be used with potential subsequent limiting if Max TP is exceeded after a positive gain offset (scenario c). This gain value can control the playout level of the file so that -23 LUFS is reached. d) The Programme Loudness Level is lower than LUFS and Loudness Range is wider than the internal tolerance for the genre or distribution channel. The Programme Loudness Level can be treated according to c above. Loudness Range is subject to processing (LRA Building Block) and thus potentially reduces Max TP. Although Max TP could have exceeded the permitted limit when applying a positive gain offset to L K, Max TP Processing might not be necessary because of the LRA reduction. A calculation of Max TP during the LRA reduction process is therefore needed. 33

34 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 e) Loudness Range is wider than the tolerance for the genre or distribution channel. As mentioned in 4.4, a compressor with low threshold and a very moderate ratio can be used to narrow LRA (Loudness Range Building Block). For files, automatic processes with a Target-LRA are advantageous. Alternatively, the result of the LRA measurement might activate a dynamics compressor preset downstream with parameters similar to those listed in 4.4. Max TP can only become lower, and so there is no potential for any True Peak Processing. f) The Maximum Permitted True Peak Level is exceeded. Exceeding the Max TP level of the respective distribution system incurs a risk of distortion further downstream (in a D-to-A-converter, sample-rate converter or bitrate reduction codec, for example). According to the Max TP Building Block True Peak Limiting is applied to lower Max TP. Whether there is a siginificant change to Programme Loudness as a result of this depends on the number and size of the peaks that are affected. Any other combination of results of the initial measurement of L K, LRA and Max TP are covered by processes already introduced in the scenarios above. 7. Metadata As described in 3.2, loudness normalisation can be either achieved through normalisation of the audio signal (the recommended method) or by using Metadata to store the actual loudness level. For the latter, the shift to Target Level can be performed either during the transfer of the audio file to the playout server, in the playout audio mixer, through choosing the appropriate preset of a downstream dynamics processor or directly at the consumer end with an adjustment of the playback level. Metadata generally can be active (potentially changing the audio signal) or descriptive (providing information about the signal, such as format, copyright etc.). As a natural consequence of the work within PLOUD and the publication of EBU R 128 and its supporting documents, the three main measures Programme Loudness, Loudness Range and Maximum True Peak Level shall form the core of loudness Metadata in audio files. Those measures are already included in the header (Broadcast Extension (BEXT) chunk) of the Broadcast Wave File (BWF) format (see EBU Tech Doc 3285 [10]; for a detailed description of BWF, see [11], [12] and [13]). Furthermore, the values for the Maximum Momentary Loudness Level as well as the Maximum Short-term Loudness Level shall be stored, as these parameters are helpful for controlling the dynamics of very short content (<30 s; see also 10). Loudness Metadata is also intended to be included in the SMPTE dictionary with potential refinements like Loudness Profiles, addressing, for example, different processing presets of downstream loudness processors. 34

35 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R 128 The Metadata parameters in existing systems that are of primary interest concerning loudness are: programme loudness dynamic range control words downmix coefficients For example, in the Dolby AC-3 Metadata system, these parameters are called dialnorm (dialogue normalisation), dynrng (dynamic range) and Centre/Surround Downmix Level. The parameter dialnorm genuinely describes the loudness of an entire programme with all its elements such as voice, music or sound effects (also a music-only programme has a dialnorm value). This may seem confusing; the reason is the focus of the Dolby system on normalisation according to the anchor signal dialogue. 7.1 Programme Loudness Metadata Following the emphasis on normalising the audio signal in production to -23 LUFS, the relevant Metadata parameter shall naturally also be set to indicate -23 LUFS, provided the programme has been normalised to the Target Level. Consequently, after widespread normalisation of the source audio signals the Programme Loudness Metadata parameter will be static. Exceptions where a different value than -23 may be used are: The programme does not fit into the window provided by -23 LUFS and -1 dbtp. This may occur mainly with very dynamic feature films and for broadcasters who want to transmit these programmes with such a large loudness-to-peak ratio; Legacy programming from the archive may not be able to be adjusted in time to fulfil the target level system of R 128; External live programmes may be provided with different loudness levels and Metadata. A fully functional system of providing and using Metadata over the whole signal chain is already in place. This implies faithful transportation of loudness Metadata to the consumer s home equipment. In all these circumstances the correct Metadata value for Programme Loudness, measured with an EBU Mode compliant meter, shall be set by all means. The distribution systems as well as Home Theatre Equipment handle this situation downstream (see EBU Tech Doc 3344). 7.2 Dynamic Range Control Metadata Just as loudness normalisation can be performed at the source audio signal or via Metadata, the same applies to dynamic range processing. In the Metadata environment, dynamic range compression information is sent as part of the datastream in the form of gain-words. In the Home Theatre Equipment of the consumer, this information is applied to reduce the dynamic range of the signal, either by default or after user activation. Dynamic range control through the use of Metadata is not comparable with a sophisticated dynamics processor, but it provides a sticking plaster for situations where the consumer wants a considerably lower dynamic range. Referring again to the Dolby Digital system, there are 6 compression presets that cause the encoder to generate different gain control words that are sent in the bitstream to the consumer's decoder: Film Standard, Film Light, Music Standard, Music Light, Speech and None. These presets result in more or less compression centred around the dialnorm value, one more reason to set this Metadata parameter correctly (see Figure 15 for the compression curves around -23 LUFS). 35

36 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Figure 15: Generic Dynamic Range compression curves of the AC-3 system Two compression profiles exist within Dolby Digital: Line mode and RF mode. For each, a separate compression preset can be chosen. Within the system of R 128 and its concept of normalising the audio signal to -23 LUFS as well as using the measure Loudness Range to determine any potential processing, the preset None may be used. This may be applicable in particular for Line mode and also by default in RF mode. The control of Loudness Range through actual processing of the audio signal at transmission is generally shifting the issue upstream. However, for specific programmes a broadcaster may choose a gentle profile for RF-mode systems (to avoid too active overload protection) while still choosing None for Line-mode systems. Broadcasters that need other profiles than None to support their internal workflow must be aware that this functionality may not always be implemented reliably in their listeners equipment. Manufacturers and distribution companies are advised to ensure that equipment is made in accordance with EBU Tech Doc 3344 ( Distribution Guidelines ). 36

37 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Downmix Coefficients These Metadata parameters (again, as an example, here for Dolby Digital) are obviously only applicable to surround sound signals, controlling the gain (in db) of the centre channel and the surround channels when mixed to Left front and Right front to derive a 2-channel-stereo signal. The loudness of a 2-ch-stereo signal, which is the result of an automatic downmix using Metadata, is dependent on: the actual downmix coefficients themselves (+3/+1.5/0/-1.5/-3/-4.5/-6/- ) the programme content in the centre and surround channels and potential safety-limiting to avoid overload Care should be taken to avoid overload of the downmixed signal. This can be achieved with a dynamics processor upstream. Static scaling (overall level reduction) should be avoided, as it systematically introduces loudness differences between the 2-channel stereo downmix and the original surround sound signal. Dynamic scaling may offer a solution. The downmix coefficients possible within the Dolby-Digital system are governed by two downmix profiles. Initially, when there was only one profile, the parameters were coarser, with -3/-4.5/-6 db for the Centre and -3/-6/- db for the Surround channels. Now, Extended Bitstream Information (Extended BSI) provides the finer intermediate steps listed above (in the first bullet point; also DVB TS downmix coefficients offer the same resolution as the Dolby Digital Extended BSI). Broadcasters should be aware of the fact that not all reproduction equipment is able to deliver the intended downmix experience if Extended BSI is used, as legacy decoders may not be able to extract this information and would fall back to the fewer and coarser coefficients of profile 1. In the case of missing or unreliable downmix Metadata, a good starting point is to look at the coefficients described in ITU-R BS [14]: L, R front: 0 db C, LS, RS: -3 db It is also again pointed out that the surround channels are weighted with +1.5 db 5 during a loudness measurement according to ITU-R BS After an automatic downmix this weighting is not applied, as the result is only frontal 2-ch-stereo (Left and Right front). Programmes with a lot of surround content will consequently exhibit potentially larger variations of the loudness of the surround mix vs. the 2-ch-stereo downmix than programmes with more conservative use of the surround channels. 5 The +1.5 db weighting coefficient for the surround signals in a loudness measurement according to ITU-R BS.1770 is not to be confused with the actual +3 db gain for the surround signals in the cinema! In the cinema, the two individual surround channels are aligned 3 db lower in level than the front channels, so that their combined level equals one front channel. The reason for this is compatibility with Mono-Surround movies (matrix-encoded Dolby Stereo has only a (bandlimited) mono-surround signal), where both surround channels would get the identical signal. For discrete multichannel audio mixes ( 5.1 etc.) the surround signals in the final mix are therefore 3 db hotter, as the mixing engineer compensates for the 3 db lower alignment of the surround channels. If a cinema mix is broadcast, this 3 db difference has to be compensated (other parameters like Loudness Range should be adapted as well). Whereas the +3 db gain for the surround channels has a purely technical reason, the +1.5 db gain for the surround signals in the loudness measurement has psychoacoustic reasons. Humans perceive sounds coming from the back louder than frontal ones with the same sound pressure level. A measurement device does not have a brain and thus needs this gain factor. 37

38 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 In any case, there can be no guarantee that the Metadata supplied with an external file (or other media) are correct. Programme Loudness Metadata indicating -27 (the factory default for dialnorm in the Dolby-Digital system) or -31 (the lowest possible value in that system) are likely to raise special awareness, as chances are that Metadata have either not been looked at or been abused for the programme to appear (much) louder when replayed at the consumer s side. It is therefore recommended to discard loudness and dynamic range control Metadata of external sources (except where the source can be fully trusted). Downmix coefficients may be passed through a fully functional Metadata system. An entire measurement process of the three main audio measures needs to be conducted afresh. Only this will ensure the correct subsequent processing. For internal purposes, Metadata can be better controlled. 8. Alignment of Signals in the Light of Loudness Normalisation 8.1 Alignment Signal and Level An Alignment Signal in broadcasting consists of a sine-wave signal at a frequency of 1 khz, which is used to technically align a sound-programme connection. In digital systems the level of such an Alignment Signal is 18 db below the maximum coding level, irrespective of the total number of bits available (-18 dbfs). The switch to loudness normalisation does NOT change this approach, as alignment does not imply a mandatory connection to loudness metering or measurement. Therefore, alignment for sound-programme exchange can be performed as usual, with a sine-wave signal of 1 khz at a level of -18 dbfs. This is specified in EBU Recommendation R 68 [15]. In the same document the Permitted Maximum Level is still mentioned, as defined in ITU-R Recommendation BS [16]; With the change to the Maximum Permitted True Peak Level (-1 dbtp for generic PCM productions) being different than the recommended -9 dbfs in ITU-R BS.645 (because of the QPPM-scenario becoming obsolete), the relevant sections of EBU R and ITU-R BS.645 (as well as documents that refer to the definition of Permitted Maximum Level within these recommendations) potentially need to be revised. The alignment level of -18 dbfs (1 khz tone) will read as -18 LUFS on a loudness meter with absolute scale (or +5 LU on the relative EBU-mode scale), provided that the 1 khz tone is present (in phase) on both the left and right channel of a stereo or surround sound signal. If the 1 khz tone is used only in a single front channel, the loudness meter will read -21 LUFS (or +2 LU on the relative scale). 38

39 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Listening Level A different topic is the Listening Level in an audio reproduction system. In the relevant document, EBU Tech Doc 3276-E Listening conditions for the assessment of sound programme material (and Supplement 1, extending it to Multichannel Sound), the following formulas are used to adjust the level of one loudspeaker [17]: (1) L LISTref = 85 10log2 db A (for 2-channel stereo) (2) L LISTref = 96 db C, referenced to digital full scale signal level (for multichannel audio up to 5.1) To achieve this, a signal consisting of noise of equal energy per octave and covering either the whole frequency range (equation (1)) or the frequency range from 500 Hz to 2 khz (equation (2)) should be employed. Measurements should actually be made at a mean signal level equal to the alignment level, which is defined here as 18 db below digital full scale. Under these conditions, the loudspeaker gains should be adjusted to achieve a Reference Listening Level (L LISTref ) of 85 3 = 82 db A Sound Pressure Level (SPL) per loudspeaker for 2-channel stereo systems and = 78 db C SPL per loudspeaker for multichannel systems. The measurements should be made at the reference listening position using an A-weighted slow response sound level meter for 2-channel stereo and a C-weighted one for multichannel audio. This is arguably a bit confusing, with different numbers, different noise signals and different frequency weighting of the sound level meter. But these differences actually compensate in a way and result in a similar listening level for both 2-channel stereo and multichannel systems. To summarize: For 2-channel stereo: L LISTref = 82 db A SPL per loudspeaker (using 20 Hz 20 khz noise of equal energy per octave at -18 dbfs rms) For 5.1 MCA: L LISTref = 78 db C SPL per loudspeaker (using 500Hz 2 khz noise of equal energy per octave at -18 dbfs rms) The abuse of peak normalisation has led to considerably lower listening levels, so that with loudness normalisation, the recommended levels might now actually be use. It is also anticipated that the average level of sound-programmes will generally be lower once EBU R 128 is put into practice. The decrease in level may be in the order of up to 3 LU (in extreme cases even more). This makes a corresponding increase in the existing monitor level of the reproduction system seem likely. As indicated above, the Alignment Level does not have to change accordingly, as the alignment procedure is still valid to ensure a reasonable gain structure as well as a high signal-to-noise ratio in the reproduction chain. If there is widespread agreement in the future on an increased level to the monitors due to the lower average level, the relevant documents will be examined again. 39

40 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 9. Implementation and Migration It is clear that such a fundamental change in the way audio signals are measured, metered and treated, and that affects all stages of audio production, distribution, archiving and transmission, is not done overnight with the flick of a switch. Every broadcaster and audio facility must find its individual way to perform this change, to install the appropriate equipment, train the staff and get on the road to loudness paradise! Nevertheless, a few constants can be stated that will be applicable for everyone. They are presented in the following sections. 9.1 Generic Migration and Implementation Advice Establish an internal loudness group to discuss basic implications and a strategy to convince management, programme makers and your colleagues. Start now don t wait until everything is in place and all the others have done it and don t try to be perfect in the very beginning. Before you can do anything, management has to agree to this change and all its consequences. Get a written agreement or call for action from the general director. Provide loudness meters to your key production personnel. Let them play with them, gain first experiences and learn the advantages and liberations of the loudness paradigm so they can be opinion leaders for their colleagues. Survey the market regarding loudness metering and loudness management to determine what is best suited for your environments. Determine the key areas in which loudness work should start. Potential candidates are: production studios, post-production suites, OB vans, QC (Quality Control) department. Be aware that you will encounter obstacles ( It has always been that way, It has never been that way, Who are you to say we should do it that way ). Patience and demonstrating practical examples will pay off. Become your facility s Zen-master of loudness normalisation ( restraint simplicity naturalness ). Allow everybody time to adapt. Although the audience has been waiting for a solution for decades, don't create more problems by trying to do too much too quickly. Solutions for file-based workflows are still rare (February 2011). Keep an eye on the market and demand solutions from vendors. Use this fundamental change as an opportunity for a general discussion about audio quality and the development of a corporate sound, which includes, for example, speech intelligibility, the balance of speech vs. music and, of course, loudness normalisation of programmes. Use and trust your ears! They are the best loudness meters. Smile when you watch your colleagues working with the loudness paradigm as if nothing else ever existed. 40

41 Tech 3343v2 Practical guidelines for Production & Implementation in accordance with R Points of Action for Migration and Implementation Establish an internal loudness group. Don t wait, start now. Get a written agreement from management. Give your key personnel loudness meters. Survey the market for loudness equipment. Choose the key areas to be the first to change. There will be obstacles. Be patient, allow time. Use the momentum to discuss audio quality. Use and trust your ears. Become your facility s Zen master of loudness. 10. Genre-specific Issues The concept of EBU R 128 centres on the loudness normalisation of each programme to one single Target Level (-23 LUFS). There are two reasons why this cannot be a perfect solution: No objective loudness measurement can ever be perfect There will always be individual preferences Thus, a perfect solution is generally not possible as loudness perception differs from person to person and is dependent on factors like age, gender, mood, etc. Within the scope of EBU R 128 it is vital to understand that it is not intended to achieve a loudness balance based on the real sound pressure level of a specific audio signal, but instead provide a satisfactory listening experience for a diverse mix of genres for the majority of listeners. This results, for example, in a Schubert string quartet having the very same integrated loudness level as a Mahler symphony, namely -23 LUFS. While this does not reflect reality, it makes these items fit into a wide array of adjacent programming, and that is the intention of advocating one single number. As this document shall serve as a pool of experiences, one might be tempted to consider refining this paradigm, once loudness normalisation becomes widespread. But listeners do accept if the loudness level of programmes lies within a so-called comfort zone of about 8-9 LU, whereas the distribution is asymmetric (for example, +3 LU/-5 LU). In cases, where the objective loudness algorithm does not always provide a perfect result, the programme will most certainly still lie within this comfort zone. Broadcasters should also bear in mind that the audience can still adjust the loudness level with their remote control, to accommodate likes and dislikes. The EBU encourages the normalisation to one single target level despite a potential refinement for individual genres. Allowing too many variations (or even only a few) may challenge the system of equal average loudness from the onset. Naturally, the fear is that variations would be biased to the louder side. 41

42 Practical guidelines for Production & Implementation in accordance with R 128 Tech 3343v2 Exhibiting a Programme Loudness Level lower than the Target Level is a slightly different topic. As a case study, two genres are now investigated, where a specific treatment (also for the Maximum Loudness Level) may be appropriate under certain circumstances; commercials (advertisements) and trailers, and music programmes Commercials (Advertisements) and Trailers This type of programme is arguably the most frequently mentioned one as far as listener annoyance is concerned, and thus is mainly responsible for the loudness problems encountered today. In the UK (BCAP rules Broadcast Committee of Advertising Practice) and the USA (CALM Act Commercial Advertisement Loudness Mitigation) as well as in Poland and Italy even legislation has recently been put into place to tame this genre. It is certainly vital that the system of loudness normalisation based on EBU R 128 provides an effective toolset for this task abuse shall be prevented. To control the dynamics of a commercial in a loudness normalised world where there exists the danger of suddenly too high loudness differences (overly loud pay-off after a longer period of low-level signals just above the gate threshold), the measure Loudness Range (LRA) is not suited, as the calculation is based on the short-term loudness values (3 s interval). Therefore, for very short items there are too few data points to derive a meaningful number for LRA. The Loudness Range parameter is not to blame for this fact, as it was never intended for this purpose. An alternative can be found in using the Maximum Momentary Loudness Level (Max ML 400 ms) and/or the Maximum Short-term Loudness Level (Max SL 3 s). Especially for very short items (<30 s), these parameters can be effectively used to limit loudness peaks. First experience of PLOUD members has pointed to a value around +8 LU (-15 LUFS) as a possible limit for Max ML and +3 LU (-20 LUFS) for Max SL. In any case, both parameters (Max ML and Max SL) are part of the extension of the Metadata of the Broadcast Wave File Format (BWF) (see EBU Tech Doc 3285, version 2.0, 2011 [10]). EBU members are encouraged to use individual limits of Max ML or Max SL for short items and report their findings. For programmes of this genre that consist of only background or creatively wanted low-level sounds, a loudness level lower than Target Level may be used. This is in line with past and current practice to limit the maximum peak level (now: loudness level), but not enforce all content to sit at that maximum. Deliberate low-level audio provides contrast, and this is one of the most fundamental creative tools in every art form. The short duration of the commercials, advertisements or trailers that might use this dramaturgical tool effectively is not likely to cause any influence on the daily long-term average loudness level of the station. Programmes destined for playout lower than Target Level need special attention to ensure they pass automatic normalisation processes unharmed. They should really be the exception, not the rule. Ultimately, the responsibility for all these cases and decisions lies with the producer, director or other creative personnel, respectively. 42